WorksPhysics and Astronomy
Wafer-to-Wafer Bonding: Part: I – The Coupled Physics Problem and the 2D Finite Element Implementation
Abstract
Wafer-to-wafer (WxW) bonding is a key enabler for three-dimensional integration, including hybrid bonding for fine-pitch Cu–Cu interconnects. During bonding, wafer deformation and the air entrapped between the wafers interact through a strongly coupled, time-dependent fluid–structure interaction (FSI) that can produce non-intuitive bonding dynamics and process sensitivities. This paper develops a mathematically consistent reduced-order model for WxW bonding by deriving a Kirchhoff–Love plate equation for wafer bending from three-dimensional linear elasticity and coupling it to a Reynolds lubrication equation for the inter-wafer air film. The resulting nonlinear plate–Reynolds system is discretized and solved monolithically in the high-performance FEniCSx framework using a interior-penalty formulation for the fourth-order plate operator, standard continuous Galerkin discretization for the pressure field, implicit time integration, and a Newton solver with automatic differentiation. Simulations reproduce experimentally reported wafer displacement histories for multiple initial gaps and verify force equilibrium at the bond front, where the Reynolds pressure acts as an effective contact reaction. Parametric studies reveal nonlinear, and in some cases non-monotonic, sensitivities of bonding-front kinetics to the initial gap, air viscosity, and interfacial energy, providing actionable trends for process optimization.
Introduction
Wafer-to-Wafer (WxW) bonding has emerged as a cornerstone technology for next-generation semiconductor packaging, enabling three-dimensional integrated circuit (3D-IC) stacking, High Bandwidth Memory (HBM), MEMS fabrication, and photonic device integration [1, 2], [3]. Unlike traditional bump-based stacking, which introduces inter-die gaps of approximately 30 m and is limited to interconnect pitches above 10 m, (hybrid) WxW bonding achieves direct Cu-to-Cu connections with pitches below 10 m, enabling substantially higher I/O density, reduced package dimensions, lower electrical resistance and superior thermal performance [1], [4]. These attributes translate directly into higher bandwidth, improved signal integrity, reduced power consumption, and enhanced heat dissipation, all critical requirements for enabling miniaturization of packages and modern HBM architectures. Despite this transformative potential, WxW bonding involves high process sensitivities, such as - submicrometer particulate contamination can nucleate voids at the bond interface, leading to delamination, incomplete bonding, high overlays, and degraded mechanical and electrical integrity. Furthermore, the governing physics of the bonding process involve a strongly coupled, nonlinear fluid–structure interaction that can give rise to non-intuitive behaviors, such as non-monotonic relationships between overlay and initial bond gap. Developing a rigorous predictive model of the bonding process is therefore not merely of academic interest but of direct industrial relevance, as it provides a pathway to understanding and mitigating these failure mechanisms.
The mechanics of wafer bonding were first analyzed using finite element methods by Turner and Spearing [5], who established the plate-bending framework for predicting bonding-induced deformations. The dynamics of the bonding front have been studied since the seminal work of Rieutord et al. [6], who proposed a model relating the bonding front velocity to the surface adhesion energy and the viscous drag of air escaping the shrinking gap between the wafers. Experimental measurements of adhesion energies and bonding wave velocities were subsequently reported by Larrey et al. [7]. This competition between elastic driving forces and viscous resistance was later formalized as a coupled Fluid–Structure Interaction (FSI) problem by Navarro et al. [8], wherein the elastic deformation of the wafer governs the air pressure between the wafers and the resulting air pressure field in turn loads the wafer. The dynamics of the bonding wave has also been analytically examined by Radisson et al. [9], who modeled the propagation speed of the bonding front as a function of the material and geometric parameters. More recently, Jain [10] derived a self-consistent analytical formulation for the bond front velocity by re-parameterizing the viscous dissipation region, providing power-law solutions for the substrate shape during bonding. More recent efforts have employed finite element methods to model various aspects of the bonding process, including bonding wave propagation and cavity-induced deformation modes [11, 12, 13], thermo-mechanical effects in glass-frit bonding [14], and Cu pad expansion in hybrid bonding [15]. Together, these studies have demonstrated the viability of the FSI framework for capturing the essential physics of WxW bonding.
Despite this body of work, existing models share three common limitations: (a) the plate equations employed are not derived from three-dimensional elasticity with explicit statement of the Kirchhoff–Love kinematic hypotheses and their validity criteria; (b) the complete weak formulations necessary for finite element implementation—including the treatment of the fourth-order plate operator, boundary conditions, and discretization details—are not provided, limiting reproducibility; and (c) the inherently nonlinear nature of the coupled system, which gives rise to non-intuitive phenomena such as non-monotonic dependence of bonding wave speed on initial gap, has not been systematically analyzed or attributed to the mathematical structure of the governing equations. Previous formulations generally rely on simplified plate models that are not rigorously derived from the full three-dimensional governing equations of linear elasticity, and they often do not clearly articulate the assumptions under which such dimensional reductions are justified [8, 11]. This gap is significant because the wafer–wafer bonding problem is inherently three-dimensional, involving aspects such as asymmetric incoming wafer geometries, the anisotropic behavior of silicon, and asymmetric boundary conditions driven by how the wafers are constrained and subsequently released. In most cases, the three-dimensional balance of linear momentum, the Kirchhoff–Love kinematic assumption, and the derivation of the flexural rigidity are neither made explicit nor tied to a properly posed boundary value problem. Furthermore, the nonlinear nature of the coupled system has been largely overlooked: the two-way coupling between wafer deflection and air pressure (between the two wafers) yields a genuinely nonlinear, time-dependent FSI problem whose solution behavior is far from intuitive. Non-intuitive phenomena—such as pressure-induced retardation of the bond front and the non-monotone dependence of bonding wave speed on initial bond gap—have not been rigorously analyzed or attributed to the structure of the governing equations.
The present work addresses these shortcomings by developing a mathematically rigorous, fully coupled FSI formulation for WxW bonding, implemented in the high-performance FEniCSx finite element framework [16, 17]. Starting from the three-dimensional equations of linear elasticity for an isotropic wafer, we perform a systematic reduction to the classical Kirchhoff–Love plate equation via through-thickness integration [18], making all assumptions and their validity criteria explicit. The air between the wafers is governed by the Reynolds lubrication equation, and the two-way coupling – wherein the wafer deflection and its rate determine the air pressure, which in turn acts as a distributed transverse load on the wafer – is preserved in full. Notably, the Reynolds pressure field in the bonded region naturally serves as a contact reaction that resists interpenetration of the wafers, eliminating the need for a separate contact algorithm. The resulting inherently nonlinear coupled equations, and the results presented in § 4 demonstrate several counterintuitive implications of this nonlinearity that have not been previously reported or discussed in the WxW bonding literature. The implementation leverages the FEniCSx framework [16, 19, 20, 21], which has proven to be effective for analogous nonlinear multiphysics problems, including nonlinear viscoelasticity [22, 23] and dielectric elastomers [24].
This paper is the first in a planned series. Part I is deliberately restricted to the reduced two-dimensional problem, which suffices for the bonding-front kinetics and air-film pressure studied here, but excludes effects that are intrinsically three-dimensional: in-plane deformation and the resulting post-bond overlay, anisotropic flexural response, asymmetric incoming wafer geometries, and lift-off of the bottom wafer from its chuck, which the present rigid, perfectly adhered bottom wafer precludes by construction. Later parts will progressively address these restrictions, with the eventual objective of solving the full three-dimensional bonding problem. Since the coupling structure established here is unchanged by such extensions, the formulation and monolithic solution strategy of Part I carry over directly; see § 5.
The remainder of this paper is organized as follows. § 2 presents the complete mathematical formulation, beginning with the three-dimensional governing equations and proceeding through their systematic reduction to the two-dimensional coupled plate – Reynolds system. § 3 describes finite element discretization and FEniCSx implementation. § 4 presents numerical results, including validation studies and the non-intuitive physical phenomena arising from the nonlinear coupling. § 5 concludes with a summary and directions for future work.
Theory
The mechanics governing Wafer-to-Wafer (WxW) bonding is characterized by a coupled Fluid-Structure Interaction (FSI) problem. Consider, two wafers in their undeformed initial configurations at time, , within a Cartesian coordinate system, positioned coaxially by the bonding apparatus on the top and bottom chuck occupying domains and respectively, such that and . The top wafer’s center is located at , and the center of the top surface of the bottom wafer is at , resulting in an initial separation distance between the bottom surface of the top wafer and the top surface of the bottom wafer. A thin layer of air is entrapped within the interstitial gap between the wafers. At this stage, it is useful to define the following regions, as illustrated in Figure 1
Figure 1. (a) Cross-sectional view (– plane) of the initial undeformed configuration. The top wafer and bottom wafer are separated by an air gap of height . Boundary surfaces and are labeled.
(b) Schematic of the wafer-to-wafer bonding process (cross-sectional view): (a) initial configuration with air gap between wafers and ; (b) a striker applies force to initiate contact at the center; (c) bond front begins to propagate radially outward from the contact zone; (d) bond front continues to advance while entrapped air is expelled; (e) the top wafer is released from the chuck and the striker is retracted; (f) full bonding across the entire interface.
Figure 1: (a) Cross-sectional view of the initial domain configuration showing the top wafer and the bottom wafer separated by an air gap of height , and (b) schematic of the six stages of the wafer-to-wafer bonding process.
Clearly, denotes the three-dimensional reference domain occupied by the top wafer. For the sake of completeness for discussions in the following sections, we introduce
as the two-dimensional mid-surface of the top wafer, obtained by projecting onto the plane .2
The subsequent bonding process encompasses the following four stages:
Initiation: The process commences when a striker of radius applies a localized force, at time , where , to the central region of the top wafer. This applied force induces a deformation in the top wafer, causing it to make initial physical contact with the bottom wafer at the point of maximum deflection and thereby modifying the effective separation between the wafers from its initial value .
Bond Propagation: Following initiation, the bond front propagates radially outward. The interfacing surfaces of both wafers undergo a pretreatment process (e.g., plasma activation) to establish attractive interfacial forces . These forces become significant when the local separation between the wafer surfaces reduces below a critical threshold such that
Consequently, these attractive forces draw the top wafer progressively towards the bottom wafer, concurrently expelling the entrapped air from the advancing bond interface. This constitutive assumption is adopted throughout the remainder of the paper.
Wafer Release: Subsequently, the top wafer is released from the retaining chuck of the bonding apparatus. This release allows the remainder of the top wafer to settle onto the bottom wafer under the influence of the propagating interfacial forces and potentially gravitational effects, facilitating the continued expansion of the bonded area.
Bond Completion: The process culminates when the bond interface has propagated across the entire surface of the wafers, resulting in a fully bonded wafer pair.
The three-dimensional governing equations
Under the assumption of a rigid bottom wafer, the deformation of the top wafer is described by the displacement field
where , denote the in-plane components and is the out-of-plane (transverse) component. The linearized strain tensor associated with (4) is
For an isotropic, linearly elastic material, the Cauchy stress tensor is given by the constitutive relation
where and are the Lamé parameters, is Young’s modulus, and is Poisson’s ratio.
Under quasi-static conditions, the balance of linear momentum for the top wafer, incorporating all body and surface forces, reads
where:
is the body-force density (e.g., ),
is the contact traction exerted by the chuck on the top surface of the bonding apparatus,
is the bonding traction (cf. (3)),
is the contact traction preventing inter-penetration of the wafers,
is the pressure of the air in the bond gap, which induces the normal load on .
The air pressure is determined by the three-dimensional Reynolds lubrication equation posed on the gap domain with gap height :
where is the dynamic viscosity of air.
The use of the Reynolds lubrication equation is justified by the following conditions, all of which are satisfied in the wafer bonding configuration [25]:
Thin-film assumption: The gap height is much smaller than the lateral extent, (assuming a large gap of ).
Low Reynolds number: Viscous forces dominate over inertial forces. Taking the bond-front velocity as the characteristic velocity, .
Quasi-steady pressure: The acoustic time scale for pressure equilibration, , is orders of magnitude smaller than the mechanical time scale (–).
Continuum validity: The Knudsen number , where denotes the molecular mean free path of the gas, classifies the flow regime: corresponds to continuum flow, to the slip-flow regime, and to the transitional regime where the continuum-based Reynolds equation breaks down [26, 27]. For air at atmospheric pressure and room temperature, . At the initial gap , , placing the flow well within the continuum regime. As the gap closes during bonding, the local
Knudsen number increases; however, the simultaneously rising air pressure near the bond front reduces (since from kinetic theory) [28], partially mitigating the increase in Kn. The regularization introduced in § 3 ensures that the effective gap remains sufficiently large to prevent the flow from entering the transitional regime.
The coupled three-dimensional FSI problem can thus be stated abstractly as:
subjected to the traction boundary conditions in Eq. (7b). Here, the coupling is two-way: the displacement determines the gap appearing in , and the pressure enters as a surface traction in . This system of governing equations shares structural similarities with models for the nonlinear viscoelasticity of rubber-like solids [22, 23] and the transient response of dielectric elastomers [24]. Such problems have been successfully implemented using the legacy FEniCS computing platform and, more recently, the high-performance FEniCSx framework [17].
Reduction to the Kirchhoff–Love plate model
Following [18], the three-dimensional problem , and the traction boundary conditions in Eq. (7b) can be reduced to a two-dimensional plate model in the thin-wafer limit . The classical Kirchhoff–Love hypothesis asserts:
Straight material fibers initially normal to the mid-surface remain straight and normal to the deformed mid-surface (i.e., the transverse shear strains vanish, , ).
The thickness of the plate does not change during deformation (i.e., ).
Kinematics. Under these hypotheses, the three-dimensional displacement field (4) reduces to the form
where are the in-plane coordinates of the mid-surface (cf. (2)) and is the transverse deflection of the mid-surface. In particular, the in-plane displacements are entirely determined by and its gradient, so that at the mid-surface ()
Substituting (12) into the strain definition (5), the only non-vanishing components of are
where is the curvature tensor of the mid-surface.
Bending moments. Invoking the plane-stress condition () appropriate for thin plates, the in-plane constitutive relation reads
The bending-moment tensor is obtained by integrating the first moment of the in-plane stress through the thickness:
| Symbol | Meaning |
| Geometric/kinematic parameters | |
| Reference (material) coordinate in ; denotes time. | |
| In-plane coordinate on the mid-surface . | |
| Three-dimensional reference domain of the top wafer. | |
| Three-dimensional reference domain of the bottom wafer. | |
| Thin interstitial air-gap domain between the wafers. | |
| Two-dimensional mid-surface domain of the top wafer (projection of onto ). | |
| Top/bottom surfaces of the top wafer; its lateral surface. | |
| Outer edge of the plate/mid-surface, . | |
| Vented portion of the boundary where the air gap is open to ambient; here . | |
| Outward unit normal. | |
| Wafer radius. | |
| Wafer thickness. | |
| Initial air-gap height (initial wafer separation). | |
| Local gap between wafers (here ). | |
| Bond-front radius, i.e. the radial position of the moving bonded/unbonded interface. | |
| Striker/pin radius. | |
| Final time of interest for . | |
| Displacement field; denotes the transverse displacement component. | |
| Linearized strain tensor. | |
| Cauchy stress tensor. | |
| Constitutive parameters | |
| Young’s modulus and Poisson’s ratio. | |
| Lamé parameters. | |
| Plate flexural rigidity, . | |
| Mass density. | |
| Dynamic viscosity of air. | |
| Force/loading parameters | |
| Body-force density (e.g., gravity, ). | |
| Gravitational acceleration. | |
| Chuck contact traction applied on . | |
| Unilateral contact traction preventing inter-penetration. | |
| Striker-applied surface traction/force used to initiate contact. | |
| Resultant transverse external loading per unit area of the mid-surface. | |
| Critical separation below which bonding forces activate. | |
| Interfacial bonding traction as a function of gap. | |
| Constant bonding traction for . | |
| Constant bonding energy . | |
| Pressure of the entrapped air in the gap. | |
| Ambient (atmospheric) pressure prescribed on . |
Table 1. Parameters appearing in § 2.
where
is the flexural rigidity of the plate.
Equilibrium. Integration of the three-dimensional balance of linear momentum (7) through the thickness, together with the traction boundary conditions on and , yields the classical Kirchhoff–Love plate equation (see, e.g., [18] and references therein):
where is the biharmonic operator and:
is the resultant transverse external loading per unit area of the mid-surface, obtained by through-thickness integration of the body force:
(Note denotes the transverse () component of the body-force density .)
is the air pressure in the gap,
is the bonding traction per unit area, and
and is the unilateral contact forces per unit area as explained in § 2.5.
Equation (18) is a fourth-order PDE in that inherits, through the through-thickness integration, all body and surface forces from the original three-dimensional governing equation (7). Note that, in these expressions, the boldface notation for the forces in (7) has been omitted to emphasize that the corresponding terms in the plate equation (18) are scalar transverse components of the forces.
In the same thin-wafer limit, the Reynolds equation (9) reduces to a two-dimensional equation on the mid-surface . Introducing the gap
with denoting the initial uniform separation, the two-dimensional pressure equation is
The reduced two-dimensional coupled FSI problem is therefore stated abstractly as:
Boundary and initial conditions
The coupled system (22)–(23) is supplemented with the following boundary and initial conditions for uniqueness of solutions. The boundary conditions described below apply to the top wafer, which is the only deformable body in the model. The bottom wafer is treated as rigid and fixed to the bottom chuck; its upper surface defines the reference plane against which contact is enforced.
Mechanical boundary conditions for the plate. On the outer edge :
where is the prescribed displacement at the wafer edge (e.g., from the top chuck constraint or the release of the top wafer from the top chuck) and a zero-slope condition is imposed for a clamped boundary.
On any symmetry boundaries (exploited in the quarter-wafer model):
Pressure boundary and initial conditions. Let denote the portion of the mid-surface boundary where the air gap between the wafers is connected to the ambient environment (i.e., the air pressure is prescribed). In the fully vented configuration considered here, we take .
where is the ambient (atmospheric) pressure and is the initial pressure distribution (typically ).
Initial condition for the deflection:
Summary of the coupled problem
For convenience, the complete coupled plate–air FSI problem is collected here. Find the transverse deflection and air pressure on the mid-surface domain for satisfying:
with the boundary conditions (21)–(26) for , the pressure conditions (23), and the initial conditions (29) and (28). The coupling is two-way: the deflection modifies the gap in and thereby the air-flow resistance, while the air pressure enters the plate equation as a distributed transverse load.
The Contacts
We model contact with the rigid chucks/wafers via (penalty-type) tractions expressed in terms of the transverse deflection .
Top-chuck contact. Let denote a penalty stiffness associated with the rigid top chuck. The corresponding contact traction is defined by
This traction penalizes (and thereby prevents) deformation states with that would violate the geometric constraint imposed by the rigid top chuck.
Bottom-wafer contact. We also introduce a unilateral contact traction to prevent interpenetration of the top wafer into the bottom wafer. In general, may be defined by an analogous penalty law activated at the closure of the gap. However, as shown in § 4.2, the Reynold’s pressure is sufficient to enforce the non-penetration constraint in the bonded regions and therefore provides an effective contact penalty. Accordingly, throughout the remainder of the paper we set
Bottom wafer–chuck contact. Contact between the bottom wafer and the bottom chuck is implicitly enforced through the rigid-wafer assumption: the bottom wafer remains stationary throughout the simulation, representing perfect adhesion to the chuck.
FEniCSx Implementation
This section details the finite element discretization and solution of the coupled plate–air system (25) using FEniCSx [16]. Exploiting the four-fold symmetry of the circular wafer, we solve on a quarter-domain with appropriate symmetry conditions. The plate equation is discretized with a interior-penalty (C0IP) Kirchhoff–Love formulation [29], and the Reynolds equation is discretized with standard continuous Galerkin finite elements. Both fields are solved monolithically.
Recap of the strong-form equations
For convenience we restate the coupled system (25). Find and satisfying:
Plate equation (from (30)):
where is the flexural rigidity (17) and the bending-moment tensor is given by (16). In operator form, , i.e.,
Reynolds equation (from (31)–(32)):
with . Here, is a small dimensionless regularization parameter; this choice is necessary to prevent from becoming unphysical as the gap closes, .
The regularization parameter maintains a minimum effective gap when full contact occurs (). This regularization has both physical and numerical justifications:
Physical basis: Even for highly polished silicon wafers, sub-micron surface roughness prevents perfect molecular contact. The root-mean-square roughness of CMP-polished wafers is typically 0.1–0.5 nm [30], but asperity peaks can extend to several nanometers. The regularized minimum gap ( for and ) is conservative relative to this physical roughness scale.
Numerical justification: The term in the Reynolds equation causes numerical ill-conditioning as . The regularization ensures that the discrete system remains well-posed and that the Newton solver converges reliably.
Continuum validity: As discussed in § 2 (condition (iv)), the Knudsen number governs the transition between continuum and rarefied gas flow regimes [26, 27]. At the regularized minimum gap (, ), the Knudsen number at atmospheric pressure is ; however, the simultaneously rising pressure near the bond front reduces (since ), keeping the flow within the slip-flow regime () where the Reynolds equation remains applicable [26].
Asymptotic validity: The mathematical validity of the Reynolds equation as a rigorous limit of the Navier–Stokes equations is established through asymptotic analysis under small aspect ratios and geometric perturbations [31]. Perturbing the film thickness by the small scaling factor confines the introduced error to , well within the accepted tolerances of the thin-film approximation.
A sensitivity analysis demonstrating that the results are insensitive to in the range [0.0025, 0.01] is presented in Appendix B.
Boundary conditions. On the boundary is decomposed into:
Symmetry edges and , where the slope vanishes, , and (cf. (26)).
Outer arc , where displacement Dirichlet conditions (cf. (24)) and the pressure venting condition (cf. (27)) are imposed.
Time discretization
The Reynolds equation (37) is the only equation containing a time derivative. We discretize it with an implicit (backward) Euler scheme. Partitioning into time steps with , and denoting , , the semi-discrete system at step reads:
Weak formulation
Let and denote the trial/test spaces for the deflection and pressure fields, respectively, and let , be test functions. Multiplying (38) by , integrating by parts twice, and multiplying (39) by with a single integration by parts, yields the following weak residuals.
Figure 2. Quarter-domain exploiting four-fold symmetry of the circular wafer mid-surface . (a) Red boundaries: dashed lines denote symmetry edges (zero normal slope); solid arc denotes the outer boundary (Dirichlet conditions). The striker region is shown near the origin. (b) The corresponding finite-element mesh (42,449 triangular elements) used in the DOLFINx computation.
Plate residual. Starting from , two integrations by parts give
where all boundary terms vanish on (essential BC) and on (symmetry, , ).
Pressure residual. Multiplying (39) by and integrating by parts:
where the boundary flux term vanishes on (, Dirichlet) and on the symmetry edges (zero normal flux by symmetry).
C0 interior-penalty spatial discretization
Since the plate weak form (40) involves second derivatives of , it requires -conforming elements. Instead, we use the interior-penalty (C0IP) method [29], which employs standard -continuous Lagrange elements and enforces inter-element -continuity weakly through penalty terms on interior facets.
Finite element spaces. Let be a conforming triangulation of and define the quadratic Lagrange spaces
and the mixed space .
Interior-penalty terms. Let denote the set of interior facets of , and for any facet let be a fixed unit normal, the facet diameter, and define the average and jump operators. The normal bending moment and normal slope on a facet are
The C0IP bending residual, replacing the continuous form (40), reads
where is the interior-penalty parameter.
On the symmetry boundaries , the condition (cf. (26)) is enforced weakly via Nitsche-type terms consistent with the C0IP formulation:
The discrete pressure residual is identical to the continuous form (41) restricted to :
Monolithic nonlinear system and Newton solver
The fully discrete monolithic residual is assembled as
At each time step , we seek such that
where incorporates homogeneous essential boundary conditions ( on , on ). The essential (Dirichlet) conditions enforced are:
on the vented outer boundary (gauge pressure),
on ,
at the striker/pin region.
Due to the nonlinear loading terms , , and the cubic gap dependence in the Reynolds equation, the discrete problem (49) is nonlinear and is solved by Newton’s method. Denoting the algebraic unknown vector and the algebraic residual vector , the Newton iteration reads
where the Jacobian is computed automatically via UFL’s [21] symbolic differentiation capabilities in FEniCSx.
The complete solution procedure is summarized in Algorithm 1.
Algorithm 1 Monolithic coupled wafer-wafer bonding simulation in FEniCSx
// -- Mesh, FE Spaces & Measures --Load Gmsh model via
gmsh.model_to_meshmesh ; build connectivity\hfill IP penalty scale
functionspace(msh, mixed_element([P2, P2]))split of trial, test, prev. step Functions
dx, dScell/interior-facet measures;ds_symtagged symmetry facets viameshtags// -- UFL Variational Forms --Plate residual (45):
F1inner(M,hess)*dxavg(M_nn)*jump*dSpenalty loadsNitsche symmetry (46):
F1 +=[-M_nn*dn_v + αD/h*dn_w*dn_v]*ds_symPressure residual (47):
F212μ(w-wn)*q*dx + Δt* h^3 *dot(grad(p),grad(q))*dxresidualF1+F2;jacobianderivative(residual, w_p, TrialFunction)[21]// -- Boundary Conditions & Solver --bcs[dirichletbc(p=0, Γ_R),dirichletbc(w=w_edge, Γ_R),dirichletbc(w=w_center, ω_S)]problemNonlinearProblem(residual, w_p, bcs, J=jacobian); PETSc LU/MUMPSOpen VTXWriter for output (BP4)
// -- Time-Stepping Loop --for do
; update
dt.value,disp_edge.value,disp_center.valuefrom CSVSolve:
problem.solve()\hfill assembles ; Newton iteration (50)w_p_n.x.array[:] = w_p.x.array[:];w_sol.interpolate(w_sub)w_vtx.write(t_{n+1});p_vtx.write(t_{n+1})\hfill write outputend for
w_vtx.close();p_vtx.close(); save radial profiles vianp.savereturn
Results
This section presents the numerical results for the wafer-bonding model; the solution workflow is outlined in Algorithm 1. We first validate the implementation against experimental top-wafer displacement measurements and illustrate the resulting quarter-symmetric displacement field (Figure 4). We then verify the force balance at the bond front by comparing the Reynolds air pressure with the computed contact reaction (§ 4.2). Next, we assess how the initial bond gap affects air-pressure build-up and the bond-front propagation speed (§ 4.3), and we conduct a parametric study on the role of air viscosity. Finally, we examine how the interfacial energy influences the bonding speed.
Figure 4. Snapshots of the displacement field at nine time instances during bonding for . The out-of-plane displacement is warped by a factor of 500 to enhance the visualization. The bottom wafer is shown in gray, and the color scale indicates the physical displacement of the top wafer.
Terminology note: Here “bond speed” denotes the radial gap-closure rate , where is the radius from the center-connected region satisfying the bonding criterion (region already bonded). It is a post-processed radial propagation rate obtained from the out-of-plane Kirchhoff–Love plate solution, rather than a material-point velocity.
Table 2 summarizes the input parameters common to all simulations presented in this section. Parameters marked with † were calibrated as described in § 4.1.
| Parameter | Symbol | Value | Source |
| Wafer radius | 150 mm | Standard 12" wafer | |
| Wafer thickness | 0.775 mm | Standard 12" wafer | |
| Young’s modulus | 148 GPa | Standard Si | |
| Poisson’s ratio | 0.27 | Standard Si | |
| Air viscosity | Pa·s | Standard (20°C) | |
| Pin (striker) radius | 5.0 mm | Assumed | |
| Initial bond gaps | 30, 70, 100 m | [11] | |
| Regularization parameter | 0.0075 | Sec. 3.1; App. B | |
| Bonding traction† | 0.03 MPa | Calibrated | |
| Bonding threshold† | 0.001 mm | Calibrated | |
| Interfacial energy† | 0.03 J/m | Calibrated | |
| Striker pressure† | 0.028 MPa | Calibrated | |
| Displacement location | 75 mm | Assumed |
Table 2. Input parameters for the wafer bonding simulations.
Validation
To validate the implementation, we performed three simulations corresponding to the bond gaps (a) , (b) , and (c) . In each case, Dirichlet boundary condition was prescribed using the displacement history reported in Figure 3 of [11]. The resulting displacement time history at the mid-region3 of the top wafer was then compared with the experimental measurements. Figure 3 shows that the model reproduces the measured response with good qualitative agreement and reasonable quantitative accuracy. Small discrepancies are expected because several inputs required for an exact match are not fully specified in [11] (e.g., interfacial energy and the precise probe location); these were therefore inferred through a calibration procedure in this work.
Figure 3. Top wafer displacement versus time at a fixed radial location for , 70, and 100 µm. Solid lines show the finite-element solution; dashed lines show the experimental measurement.
The parameters not specified in [11] are the striker pressure , the bonding traction magnitude (Eq. (3)), the bonding activation threshold , and the radial location of the displacement sensor. The sensor location was assumed to be mm based on the reported experimental setup. The striker pressure, the bonding traction and threshold were calibrated by minimizing the discrepancy between the simulated and measured displacement time histories at the sensor location. The calibrated values are MPa and mm ( µm). The product J/m corresponds to the interfacial bonding energy, which falls within the range of 20–100 mJ/m reported for room-temperature hydrophilic silicon wafer bonding [3, 7]. Furthermore, since the wafer release condition was not reported in [11], the measured edge displacement time history was instead prescribed as a displacement boundary condition on the edge . The striker pressure, applied uniformly over the central region , was calibrated by matching the measured wafer-center displacement history, giving MPa.
The effect of air pressure is evident in both the experiments and the simulations. In contrast to what is expected for bonding in the absence of air (pure-plate-bending), in the presence of air, the largest initial bond gap reaches complete bonding earliest, whereas the smallest gap reaches it last. This ordering is also evident in the bonded-radius evolution discussed in § 4.3.
Snapshots of the simulated out-of-plane displacement field are shown as 2D contour plots at selected time instants in Figure 4.
Reynold's pressure acting as a contact force
As discussed in § 2.5, we set the bottom-wafer contact tractions to zero, , because the air-film pressure obtained from the Reynolds solution, , is sufficient to enforce non-penetration contact. This can be seen from Eq. (37): the Reynolds equation is solved over the full computational domain , including regions that are already bonded. In the bonded region the Reynolds pressure acts as a reaction pressure, i.e., an effective contact traction that prevents interpenetration of the wafers. This is because of the inherent mathematical structure of the Reynold’s equation in which in Eq. (37) cannot be negative. Figure 5 confirms this interpretation by showing that the Reynolds pressure behind the bond front balances the applied striker pressure. So, at the bond front, the total Reynold’s pressure is:
Figure 5. Displacement (left) and pressure (right) at the final time s for m. The average pressure is computed by integrating the Reynolds pressure over ; the resulting average pressure is MPa.
where, is the actual pressure due to the air between the two wafers in the unbonded region and is the contact pressure.
We note that the Reynolds pressure computed in the bonded region (where the effective gap approaches ) does not represent physical air pressure, as the air has been expelled from bonded areas. Rather, this pressure serves as an effective contact reaction that enforces the non-penetration constraint, analogous to a penalty-regularized contact formulation. This interpretation is validated by the force balance shown in Figure 5. The regularization parameter (cf. § 3.1) is essential for this contact-reaction interpretation: by maintaining a minimum effective gap, it prevents the pressure singularity that would occur as in the unregularized Reynolds equation while allowing the Reynolds equation to return a finite, well-defined pressure in bonded regions.
Effect of bond gap on bond speed
In this section, we examine the bonding speed for the three bond gaps considered in the validation exercise in § 4.1. As shown in Figure 6, the response is nonlinear and non-monotonic. Unlike bonding in the absence of air (pure plate bending), the largest bond gap reaches complete bonding earlier than the smallest gap; see Figure 6(a). This ordering is driven by the air-film pressure in the unbonded region: smaller gaps generate higher , which increases the resistance to bond-front advance as in demonstrated in Figure 10b. The instantaneous bond-front speed remains nonlinear and is not ordered monotonically for all times, as illustrated in Figure 6(b) and Figure 6(c). This behavior is expected, since the bond-front speed and the evolution of air pressure are governed by Reynolds’ equation. Additional validation plots for the spatio-temporal displacement fields and air-pressure diagnostics are provided in Appendix A; see Figure 9 and Figure 10.
Figure 6. Effect of the initial bond gap for , , and : (a) bonded radius versus time, (b) bond speed versus time, and (c) bond speed versus radial position. Before bond initiation, and .
Figure 9. Spatio-temporal evolution of the radial displacement field for three bond gaps.
Figure 10. Air-pressure diagnostics for , , and for the result reported in § 4.3: (a) radial profiles in the unbonded region at ; and (b) profiles at time when . The inset in (b) magnifies the bond-front region .
Effect of Air Viscosity on air pressure and Bond Speed
Varying the air viscosity modifies the Reynolds pressure field and, consequently, the bond-front propagation rate. Figure 7 summarises the post-processed diagnostics used to quantify this effect: bond speed versus time, the spatial variation of bond speed, and the temporal evolution of the bonded radius. For the same time instant, lower viscosity yields a faster bond front and a larger bonded area.
Figure 7. Effect of air viscosity for and , 0.50, 0.75, and 1.00: (a) bond speed over time, (b) bond speed versus radius, and (c) bonded-radius evolution, where is the reference air viscosity.
Effect of Interfacial Energy on bond speed
Varying the interfacial energy changes the thermodynamic driving force for bonding, which in turn affects the bond speed. Figure 8 presents the diagnostics that capture this sensitivity: the bond speed as a function of time, the spatial distribution of the bond speed, and the temporal evolution of the bonded radius.
Figure 8. Effect of interfacial energy for and , 0.010, 0.020, and 0.030 MPa, corresponding to , 0.010, 0.020, and 0.030 J : (a) bond speed over time, (b) bond speed versus radius, and (c) bonded-radius evolution.
Conclusion and Future Work
Conclusions
This work developed a mathematically consistent reduced-order model for wafer-to-wafer bonding by coupling a Kirchhoff–Love plate description of wafer bending to a Reynolds lubrication model for the entrapped air film. Starting from three-dimensional linear elasticity, the plate equation was obtained in the thin-wafer limit and combined with a pressure evolution law in which the wafer deflection directly modulates the local gap height and flow resistance. The resulting nonlinear, strongly coupled system was discretized in a monolithic manner and implemented in the FEniCSx framework using a interior-penalty formulation for the fourth-order plate operator together with standard continuous Galerkin elements for the pressure field. The numerical studies demonstrated consistent force balance in the contact region and produced bonding-front dynamics in agreement with experimentally measured wafer-displacement histories, thereby providing a predictive and computationally efficient tool for assessing process sensitivities.
A key advantage of the present formulation is that the coupling structure is explicit: the air pressure enters the plate equation as a distributed transverse load, while the plate deflection enters the Reynolds equation through the evolving gap . This structure enables systematic extensions of the physics without altering the overall solution strategy.
Extension to anisotropic flexural response
In the current isotropic plate model, the bending stiffness enters through the scalar flexural rigidity . For crystalline wafers and layered stacks, however, the effective bending response can be direction-dependent. Within the same Kirchhoff–Love framework, anisotropy can be incorporated by replacing the scalar stiffness with a bending-stiffness tensor. In practice, one introduces the bending moment–curvature relation
where and are the bending-moment and curvature tensors, respectively, and is a symmetric fourth-order bending stiffness (often expressed, under plane-stress reduction, as a matrix in Voigt notation). The plate equilibrium then becomes
which reduces to when corresponds to isotropic bending. Because the Reynolds equation and the coupling through remain unchanged, the anisotropic extension primarily modifies the plate operator and the constitutive evaluation of bending moments, and can be integrated into the present monolithic Newton framework with minimal changes.
Future work
Several extensions are planned. First, we will incorporate anisotropic and heterogeneous bending stiffness to represent crystallographic effects, device layers, and patterned stacks, enabling more accurate prediction of bonding kinetics for modern wafer architectures. Second, we will extend the reduced two-dimensional plate–Reynolds formulation to a geometrically richer shell model. A shell description will allow simultaneous prediction of (i) bond-front propagation and pressure evolution in the interfacial air film and (ii) in-plane deformations and post-bond overlays, including wafer stretch, shear, and in-plane distortion. This, in turn, will enable prediction of post-bond wafer shapes and residual distortions (e.g., global bow/warp and in-plane misfit patterns) that are critical for overlay, alignment, and downstream process yield. Finally, we will investigate the effect on bonding due to incoming wafer geometries.
Additional validation results
This appendix section collects additional plots supporting the validation study in § 4.1. Figure 9 presents the full spatio-temporal displacement field for each initial gap, from which the bond-front trajectories reported in Figure 6(a) are extracted. Figure 10 presents the radial air-pressure profiles that explain the effect of bond gap on bond speed as discussed in § 4.3.
Sensitivity to regularization parameter
This appendix presents the sensitivity of the simulation results to the regularization parameter introduced in § 3.1. Simulations were performed for with . Figure 11 demonstrates that the displacement and pressure are approximately insensitive to within this range, confirming that the choice used throughout this work does not materially affect the reported results.
Figure 11. Effect of regularization parameter on successive difference norms for : (a) normalized deflection difference versus time, and (b) normalized pressure difference versus time, for consecutive pairs from .
References
- [1]Shuyu Bao, Yue Wang, Khaw Lina, Li Zhang, Bing Wang, Wardhana Aji Sasangka, Kenneth Eng Kian Lee, Soo Jin Chua, Jurgen Michel, Eugene Fitzgerald, Chuan Seng Tan, and Kwang Hong Lee. A review of silicon-based wafer bonding processes, an approach to realize the monolithic integration of Si-CMOS and III–V-on-Si wafers. Journal of Semiconductors, 42 (2):023106, 2021. doi: 10.1088/1674-4926/42/2/023106.DOI
- [2]H. Moriceau, F. Rieutord, F. Fournel, Y. Le Tiec, L. Di Cioccio, C. Morales, A. M. Charvet, and C. Deguet. Overview of recent direct wafer bonding advances and applications. Advances in Natural Sciences: Nanoscience and Nanotechnology, 1(4):043004, 2011. doi: 10.1088/2043-6262/1/4/043004.DOI
- [3]U. Gösele and Q.-Y. Tong. Semiconductor wafer bonding. Annual Review of Materials Science, 28:215–241, 1998. doi: 10.1146/annurev.matsci.28.1.215.DOI
- [4]L. Wang, G. Fountain, B. Lee, G. Gao, C. Uzoh, S. McGrath, P. Enquist, S. Arkalgud, and L. Mirkarimi. Direct bond interconnect (DBI) for fine-pitch bonding in 3D and 2.5D integrated circuits. 2023 IEEE 73rd Electronic Components and Technology Conference (ECTC), pages 1–8, 2023. doi: 10.37665/ppobdcm64121.
- [5]K. T. Turner and S. M. Spearing. Modeling of direct wafer bonding: Effect of wafer bow and etch patterns. Journal of Applied Physics, 92(12):7658–7666, 2002. doi: 10.1063/1.1521792.DOI
- [6]F. Rieutord, B. Bataillou, and H. Moriceau. Dynamics of a bonding front. Physical Review Letters, 94:236101, 2005. doi: 10.1103/PhysRevLett.94.236101.DOI
- [7]V. Larrey, G. Mauguen, F. Fournel, D. Radisson, F. Rieutord, C. Morales, C. Bridoux, and H. Moriceau. Adhesion energy and bonding wave velocity measurements. ECS Transactions, 75(9):145–152, 2016. doi: 10.1149/07509.0145ecst.DOI
- [8]E. Navarro, Y. Bréchet, R. Moreau, T. Pardoen, J.-P. Raskin, O. Barthelemy, and I. Radu. Direct silicon bonding dynamics: a coupled fluid/structure analysis. Applied Physics Letters, 103:034104, 2013. doi: 10.1063/1.4813312.DOI
- [9]D. Radisson, F. Fournel, and E. Charlaix. Modelling of the direct bonding wave. Microsystem Technologies, 21:969–971, 2015. doi: 10.1007/s00542-015-2445-3.DOI
- [10]Utkarsh Jain. Bond front velocity in lubrication-mediated bonding of flexible substrates. Proceedings of the Royal Society A, 482:20260071, 2026. doi: 10.1098/rspa.2026.0071.DOI
- [11]Nathan Ip, Nima Nejadsadeghi, Carlos Fonseca, Norifumi Kohama, and Kimio Motoda. Multi-physics simulation of wafer-to-wafer bonding dynamics. In 2022 IEEE 72nd Electronic Components and Technology Conference (ECTC), pages 502–506. IEEE, 2022. doi: 10.1109/ECTC51906.2022.00085.DOI
- [12]Kyeongbin Lim et al. Design and simulation of symmetric wafer-to-wafer bonding compensating a gravity effect. In 2020 IEEE 70th Electronic Components and Technology Conference (ECTC), pages 1480–1485. IEEE, 2020. doi: 10.1109/ECTC32862.2020.00234.DOI
- [13]Feixiang Tang, Siyu He, Yuhan Li, Wenjin Liu, Fang Dong, and Sheng Liu. Numerical and mechanical analysis of direct wafer bonding considering non-uniform impurity particle distributions. Microsystems & Nanoengineering, 11:151, 2025. doi: 10.1038/s41378-025-00994-4.DOI
- [14]Seyed Amir Fouad Farshchi Yazdi, Matteo Garavaglia, Aldo Ghisi, and Alberto Corigliano. An experimental and numerical study on glass frit wafer-to-wafer bonding. Micromachines, 14(1):165, 2023. doi: 10.3390/mi14010165.
- [15]Yan Wen Tsau, Joke De Messemaeker, Abdellah Salahouelhadj, Mario Gonzalez, Liesbeth Witters, Boyao Zhang, Marc Seefeldt, Eric Beyne, and Ingrid De Wolf. Simulation of Cu pad expansion in wafer-to-wafer Cu/SiCN hybrid bonding. Microelectronics Reliability, 138:114716, 2022. doi: 10.1016/j.microrel.2022.114716.DOI
- [16]Igor A Baratta, Joseph P Dean, Jørgen S Dokken, Michal Habera, Jack Hale, Chris N Richardson, Marie E Rognes, Matthew W Scroggs, Nathan Sime, and Garth N Wells. DOLFINx: the next generation FEniCS problem solving environment. Zenodo, 2023. doi: 10.5281/zenodo.10447666. URL https://doi.org/10.5281/zenodo.10447666. Preprint.
- [17]Kamalendu Ghosh and Bhavesh Shrimali. A hybrid finite element implementation of two-potential constitutive model of dielectric elastomers. Finite Elements in Analysis and Design, 247:104348, 2025. doi: 10.1016/j.finel.2025.104348.DOI
- [18]Bhavesh Shrimali, Matteo Pezzulla, Samuel Poincloux, Pedro M Reis, and Oscar Lopez-Pamies. The remarkable bending properties of perforated plates. Journal of the Mechanics and Physics of Solids, 154:104514, 2021. doi: 10.1016/j.jmps.2021.104514.DOI
- [19]Matthew W Scroggs, Jørgen S Dokken, Chris N Richardson, and Garth N Wells. Construction of arbitrary order finite element degree-of-freedom maps on polygonal and polyhedral cell meshes. ACM Transactions on Mathematical Software, 48(2):1–23, 2022.DOI
- [20]Matthew W Scroggs, Igor A Baratta, Chris N Richardson, and Garth N Wells. Basics: a runtime finite element basis evaluation library. Journal of Open Source Software, 7(73):3982, 2022.DOI
- [21]Martin S Alnæs, Anders Logg, Kristian B Ølgaard, Marie E Rognes, and Garth N Wells. Unified form language: A domain-specific language for weak formulations of partial differential equations. ACM Transactions on Mathematical Software, 40(2):1–37, 2014.
- [22]Kamalendu Ghosh, Bhavesh Shrimali, Aditya Kumar, and Oscar Lopez-Pamies. The nonlinear viscoelastic response of suspensions of rigid inclusions in rubber: I—Gaussian rubber with constant viscosity. Journal of the Mechanics and Physics of Solids, 154:104544, 2021. doi: 10.1016/j.jmps.2021.104544.DOI
- [23]Bhavesh Shrimali, Kamalendu Ghosh, and Oscar Lopez-Pamies. The nonlinear viscoelastic response of suspensions of vacuous bubbles in rubber: I—Gaussian rubber with constant viscosity. Journal of Elasticity, 153(4):479–508, 2023. doi: 10.1007/s10659-021-09868-y.
- [24]Kamalendu Ghosh and Oscar Lopez-Pamies. On the two-potential constitutive modeling of dielectric elastomers. Meccanica, 56(6):1505–1521, 2021. doi: 10.1007/s11012-020-01179-1.DOI
- [25]Bernard J. Hamrock, Steven R. Schmid, and Bo O. Jacobson. Fundamentals of Fluid Film Lubrication. Marcel Dekker, New York, 2nd edition, 2004.DOI
- [26]Albert Burgdorfer. The Influence of the Molecular Mean Free Path on the Performance of Hydrodynamic Gas Lubricated Bearings. Journal of Basic Engineering, 81(1):94–98, 1959. doi: 10.1115/1.4008375.DOI
- [27]Shigehisa Fukui and Raj Kaneko. Analysis of Ultra-Thin Gas Film Lubrication Based on Linearized Boltzmann Equation: First Report—Derivation of a Generalized Lubrication Equation Including Thermal Creep Flow. Journal of Tribology, 110(2):253–261, 1988. doi: 10.1115/1.3261594.DOI
- [28]Etienne Navarro. Direct Wafer Bonding Dynamics. PhD thesis, Université de Grenoble, 2014. HAL Id: tel-01048574.DOI
- [29]Tom Gustafsson, Rolf Stenberg, and Juha Videman. On Nitsche’s method for elastic contact problems. SIAM Journal on Scientific Computing, 42(2):B425–B446, 2020. doi: 10.1137/19M1246869.DOI
- [30]C. Teichert, J. F. MacKay, D. E. Savage, M. G. Lagally, M. Brohl, and P. Wagner. Comparison of surface roughness of polished silicon wafers measured by light scattering topography, soft-x-ray scattering, and atomic-force microscopy. Applied Physics Letters, 66(18):2346–2348, 1995. doi: 10.1063/1.113978.DOI
- [31]Guy Bayada and Michèle Chambat. The Transition Between the Stokes Equations and the Reynolds Equation: A Mathematical Proof. Applied Mathematics and Optimization, 14(1):73–93, 1986. doi: 10.1007/BF01442229.DOI

